Re: [Libreoffice] minutes of tech. steering call ...

On Thu, 24 Nov 2011 20:23:22 +0000
Caolán McNamara <caolanm@redhat.com> wrote:

> On Thu, 2011-11-24 at 16:26 +0000, Michael Meeks wrote:
> >     + Bjoern has a nice write-up of how to debug it here:
> i.e. sommat like
> a) just move that out of dev-tools into bin or solenv/bin
> b) tweak it to honour $COLORTERM/$TERM -e
> c) add a handle SIGPIPE nostop noprint pass
> d) get the definitely correct pid in there
> ---
> e) get the tooling to catch the disconnected error and say that
> LibreOffice crashed, rather than the current somewhat obscure error
> f) spit out a line liner to rerun the first failing test in isolation
> with soffice server under gdb

Hmm, I just had another crazy idea, since we are mostly interested in
two things from the junit test:
- does the product work as expected?
  that works reasonably well as long as the test doesnt crash
  however crashes are icky to detect reliable "from the outside"
- if it crashes, we want to have a backtrace

However, we are not so much interested in interactively working with
soffice in the subsequenttest. So how about a very old fashioned and
almost forgotten way to debug: creating a core dump!

For bonus points one could then even print the stacktrace of a crashed
test right from "make subsequentcheck".

Opinions?

Best,

Bjoern

-- 
https://launchpad.net/~bjoern-michaelsen